ddl: retry prepare RPC when meets region error

ddl/BUILD.bazel

@@ -120,6 +120,7 @@ go_library(
         "@com_github_ngaut_pools//:pools",
         "@com_github_pingcap_errors//:errors",
         "@com_github_pingcap_failpoint//:failpoint",
+        "@com_github_pingcap_kvproto//pkg/errorpb",
         "@com_github_pingcap_kvproto//pkg/kvrpcpb",
         "@com_github_pingcap_kvproto//pkg/metapb",
         "@com_github_pingcap_log//:log",

ddl/cluster.go

@@ -24,6 +24,7 @@ import (
 
 	"github.com/pingcap/errors"
 	"github.com/pingcap/failpoint"
+	"github.com/pingcap/kvproto/pkg/errorpb"
 	"github.com/pingcap/kvproto/pkg/kvrpcpb"
 	"github.com/pingcap/tidb/ddl/util"
 	"github.com/pingcap/tidb/domain/infosync"
@@ -324,15 +325,36 @@ func SendPrepareFlashbackToVersionRPC(
 		if err != nil {
 			return taskStat, err
 		}
+		failpoint.Inject("mockPrepareMeetsEpochNotMatch", func(val failpoint.Value) {
+			if val.(bool) && bo.ErrorsNum() == 0 {
+				regionErr = &errorpb.Error{
+					Message:       "stale epoch",
+					EpochNotMatch: &errorpb.EpochNotMatch{},
+				}
+			}
+		})
 		if regionErr != nil {
-			return taskStat, errors.Errorf(regionErr.String())
+			err = bo.Backoff(tikv.BoRegionMiss(), errors.New(regionErr.String()))
+			if err != nil {
+				return taskStat, err
+			}
+			continue
 		}
 		if resp.Resp == nil {
-			return taskStat, errors.Errorf("prepare flashback missing resp body")
+			logutil.BgLogger().Warn("prepare flashback miss resp body", zap.Uint64("region_id", loc.Region.GetID()))
+			err = bo.Backoff(tikv.BoTiKVRPC(), errors.New("prepare flashback rpc miss resp body"))
+			if err != nil {
+				return taskStat, err
+			}
+			continue
 		}
 		prepareFlashbackToVersionResp := resp.Resp.(*kvrpcpb.PrepareFlashbackToVersionResponse)
 		if err := prepareFlashbackToVersionResp.GetError(); err != "" {
-			return taskStat, errors.Errorf(err)
+			boErr := bo.Backoff(tikv.BoTiKVRPC(), errors.New(err))
+			if boErr != nil {
+				return taskStat, boErr
+			}
+			continue
 		}
 		taskStat.CompletedRegions++
 		if isLast {

tests/realtikvtest/brietest/BUILD.bazel

@@ -14,6 +14,7 @@ go_test(
     deps = [
         "//config",
         "//ddl/util",
+        "//parser/model",
         "//parser/mysql",
         "//sessionctx/binloginfo",
         "//store/mockstore/mockcopr",

tests/realtikvtest/brietest/flashback_test.go

@@ -22,6 +22,7 @@ import (
 
 	"github.com/pingcap/failpoint"
 	ddlutil "github.com/pingcap/tidb/ddl/util"
+	"github.com/pingcap/tidb/parser/model"
 	"github.com/pingcap/tidb/testkit"
 	"github.com/pingcap/tidb/tests/realtikvtest"
 	"github.com/stretchr/testify/require"
@@ -90,3 +91,48 @@ func TestFlashback(t *testing.T) {
 		require.NoError(t, failpoint.Disable("tikvclient/injectSafeTS"))
 	}
 }
+
+func TestPrepareFlashbackFailed(t *testing.T) {
+	if *realtikvtest.WithRealTiKV {
+		store := realtikvtest.CreateMockStoreAndSetup(t)
+
+		tk := testkit.NewTestKit(t, store)
+
+		timeBeforeDrop, _, safePointSQL, resetGC := MockGC(tk)
+		defer resetGC()
+
+		tk.MustExec(fmt.Sprintf(safePointSQL, timeBeforeDrop))
+		tk.MustExec("use test")
+		tk.MustExec("drop table if exists t")
+		tk.MustExec("create table t(a int, index i(a))")
+		tk.MustExec("insert t values (1), (2), (3)")
+
+		time.Sleep(1 * time.Second)
+
+		ts, err := tk.Session().GetStore().GetOracle().GetTimestamp(context.Background(), &oracle.Option{})
+		require.NoError(t, err)
+
+		injectSafeTS := oracle.GoTimeToTS(oracle.GetTimeFromTS(ts).Add(100 * time.Second))
+		require.NoError(t, failpoint.Enable("tikvclient/injectSafeTS",
+			fmt.Sprintf("return(%v)", injectSafeTS)))
+		require.NoError(t, failpoint.Enable("github.com/pingcap/tidb/expression/injectSafeTS",
+			fmt.Sprintf("return(%v)", injectSafeTS)))
+		require.NoError(t, failpoint.Enable("github.com/pingcap/tidb/ddl/mockPrepareMeetsEpochNotMatch", `return(true)`))
+
+		tk.MustExec("insert t values (4), (5), (6)")
+		tk.MustExec(fmt.Sprintf("flashback cluster to timestamp '%s'", oracle.GetTimeFromTS(ts)))
+
+		tk.MustExec("admin check table t")
+		require.Equal(t, tk.MustQuery("select max(a) from t").Rows()[0][0], "3")
+		require.Equal(t, tk.MustQuery("select max(a) from t use index(i)").Rows()[0][0], "3")
+
+		jobMeta := tk.MustQuery("select job_meta from mysql.tidb_ddl_history order by job_id desc limit 1").Rows()[0][0].(string)
+		job := model.Job{}
+		require.NoError(t, job.Decode([]byte(jobMeta)))
+		require.Equal(t, job.ErrorCount, int64(0))
+
+		require.NoError(t, failpoint.Disable("github.com/pingcap/tidb/expression/injectSafeTS"))
+		require.NoError(t, failpoint.Disable("tikvclient/injectSafeTS"))
+		require.NoError(t, failpoint.Disable("github.com/pingcap/tidb/ddl/mockPrepareMeetsEpochNotMatch"))
+	}
+}